Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Render3DPass

Depth-tested + MSAA-aware render pass. The rate-limiting risk ticket of the whole wisp-3d rollout — wgpu validates depth + MSAA + pipeline state at DRAW time, not at pipeline creation, so mismatches are silent until you submit.

The MSAA-sample-count trap

wgpu requires THREE values to match:

  1. The depth texture's sample_count in its TextureDescriptor.
  2. The pipeline's multisample.count field.
  3. The COLOR attachment's sample_count (the surface texture's view).

If any pair disagrees, the failure mode is a Validation Error / Pipeline ... is bound with sample count X at submit time — NOT at pipeline creation, and NOT at attachment bind. The plumbing here keeps all three in lock-step via the single msaa_samples constructor argument; do not split this knob across multiple knobs.

Per-frame flow

sequenceDiagram
  autonumber
  participant App as wisp::Application
  participant Pass as Render3DPass
  participant GPU as wgpu Encoder
  App->>Pass: draw(encoder, color_view, camera, meshes, clear)
  Pass->>GPU: queue.write_buffer(view_proj)
  loop per mesh
    Pass->>GPU: create_buffer(vbuf + ibuf + model_ubo)
    Pass->>GPU: queue.write_buffer × 3
    Pass->>GPU: create_bind_group(model_bg)
  end
  Pass->>GPU: begin_render_pass(color + depth)
  Pass->>GPU: set_pipeline + set_bind_group × 2
  loop per mesh
    Pass->>GPU: set_vertex_buffer + set_index_buffer
    Pass->>GPU: draw_indexed
  end
  Pass-->>App: end_render_pass (RAII)

What's deferred

  • Instancing. Today each mesh becomes its own VBO+IBO+UBO write per draw. The path is open: collapse meshes that share (Mesh3D, Material) into one draw with a per-instance buffer of Mat4 + tint. Lands when scale demands it.
  • Pipeline cache. The default pipeline is one-per-pass. The Material3D trait (next chapter) introduces a TypeId-keyed cache for user shaders.